GOAT Index Boxing — how the rankings are made

Fighter information comes from Wikidata, and records from the English Wikipedia bout tables (Professional boxing record). Everything is fetched through the APIs those sites provide; no pages are scraped. Wikipedia is CC BY-SA 4.0 and Wikidata is CC0, and this site publishes an adapted form of that material. Every fighter page carries a link to the source article.

Only the English edition is read for bout tables. Other language editions are structured differently and are not supported at present. Coverage reaches back to bouts in the early twentieth century, and takes in every bout listed in the table, not only title fights.

Results from every weight class are treated as one large network rather than being split by division. If A beat B and B beat C, that is a handle on comparing A and C even though they never met. Three things go into the assessment.

1. The strength of the opponent — the stronger the fighter you beat, the further it moves
2. How the win came — a KO, a TKO or a corner retirement counts for more the earlier it lands, and a split decision counts for less than a unanimous one
3. How certain it is — a fighter with few bouts is treated as having a wide margin of error

How a win came affects only the size of the score, never how the certainty moves. A spectacular win does not make a bout better understood. A draw counts half to each fighter, and a no contest is left out of the calculation altogether. A disqualification counts as a win and a loss, but is treated as neutral as to how the win came. Announced bouts are not results, so they are not in the calculation.

Because the whole network is solved as a single answer, the result of one bout moves not only that fighter's assessment but those of the fighters around them. Men and women almost never meet, so in practice they form separate networks, and the rankings are separated by sex too.

The following are not considered at all: build, reach and height, age, stylistic match-ups, condition and injury, weight cutting, venue or home advantage, betting odds, and era adjustment. Of course these affect results, but none of them can be quantified consistently from public data, so they are deliberately left out.

1. Rating — an estimate of present strength
2. All-time (peak) — the highest rating a fighter ever held, less the certainty as it stood at that moment. This is the important point. Certainty decays as time passes since the last bout, so subtracting today's certainty would push a fighter down for no reason other than being from an earlier era. The board is meant to rank fighters at their best, and what deserves the penalty is a thin record, not a distant one. So the figure used is how certain the rating was when the peak was set
3. Active P4P — active fighters only, in rating order
4. Movement — the difference from the rank at the previous publication, compared within each view against that view's previous rank

Fighters with 10 or more career bouts are listed. Champions are exempt from that condition. A fighter counts as active if the year of their last bout falls within the most recent 2 years. Retirement is not decided automatically; it is applied by hand where it has been confirmed. Retired fighters stay on the all-time rankings.

A divisional ranking puts that division's champion at the top and orders the rest by rating, so the numbers do not always descend from top to bottom. A fighter whose sex could not be confirmed appears on no ranking at all.

An update runs in the order fetch, recompute, check, publish, and anything that fails the check is not published. Keeping broken data off the site matters more here than freshness. The update runs automatically every Sunday at 23:30 JST. A week that fails the checks is not published, and the previous week's pages stay up. Every update recomputes the ratings for every fighter from scratch.

Plainly stated:

1. It centres on fighters who have a bout table on the English Wikipedia. A fighter without one still enters the network as somebody's opponent, but their own assessment has to be built from the other side's record, so its certainty is low and they rarely surface near the top
2. Records depend on what the articles say. An error in an article is reproduced here
3. For fighters of the first half of the twentieth century, opponents' records are often incomplete, and the estimate can come out lower than the truth. Using the certainty as it stood at the peak removes the penalty for simply being old, but the disadvantage of a thin record itself remains
4. Women's careers hold fewer bouts, and even highly placed fighters can have a short record. Read the numbers as carrying a wider margin than the men's
5. Non-title bouts are included on the same footing. Eliminators and comeback fights are not distinguished
6. Updating is not automated. A recent bout can take a while to appear

Please get in touch through the contact form. A source reference makes it certain. Records themselves are read from what Wikipedia says, so once the article is right the next update picks it up automatically. Display names are corrected directly, in an override table held on this site.